Try 2, much better than before and now works fine at 0.7 strength and default cfg.
Found out you can upload more training images on the web version than the app so it should understand the concept of soiling better, and you can make the base model anything v5 which lessens the impact of the cartoonier styles of references and greatly improved the model.
Trigger words are similar to before: bulge, soiling, pantypoop, messing. Removed the word 'panties' as a trigger so it interferes with the rest of the prompt less.
I also can't see how to make this the 'new version' of V1 as opposed to a separate Lora but I guess you can just choose either for now.
